Skip to main content
Dryad

Data from: Mapping coastal redwoods (Sequoia sempervirens) across their natural range: An updateable and field-validated distribution map using Sentinel satellite data and cloud computing

Banerjee, Subham1; Francis, Emily1; Mast, Colin1

Published Jan 05, 2026 on Dryad. https://doi.org/10.5061/dryad.34tmpg4xf

Data files

Jan 05, 2026 version files 174.03 MB

Click names to download individual files

Abstract

Coast redwood (Sequoia sempervirens) is a uniquely-tall and long-lived tree species that occupies a narrow fog-belt along the Pacific coast of North America. Despite its ecological and conservation significance, existing maps of redwood distribution remain limited in spatial resolution, accuracy, and timeliness. In this study, we present an updatable and field-validated distribution map of S. sempervirens across its entire native range – from southwestern Oregon to central California – developed using freely available Sentinel-2 multispectral and Sentinel-1 SAR data. We compiled a georeferenced canopy classification dataset of 960 points, combining field surveys conducted in October 2024 with field-based ground truth points collected in 2017 from a prior mapping study, and externally sourced field-based redwood presence records. This dataset was used to train machine learning models (Random Forest and Gradient Boosted Trees) within a cloud computing framework to classify redwood presence and absence at 10 m spatial resolution. Binary classification models achieved high predictive performance, with the best model yielding over 88% overall accuracy and an AUC of 0.92 on a 30% hold-out validation set. Ten-fold cross-validation on the training data further confirmed model consistency, with high true positive rates and low false positive rates across folds. A secondary multi-class model differentiated between redwood-dominated and mixed-conifer forest types, achieving an overall accuracy of 73.82%. Comparison with previous redwood distribution datasets revealed substantial agreement but also significant discrepancies, with the new model suggesting redwood presences in previously unmapped redwood fragments and absences in locations mapped as redwood. Validation against field data confirmed higher accuracy in the new map. The resulting range-wide redwood map offers a current, accurate, and updateable platform for conservation planning, habitat monitoring, and ecological research. It also establishes a high-confidence baseline for tracking redwood distribution dynamics under ongoing climate and land-use change.